Whither lung EIT: Where are we, where do we want to go and what do we need to get there?

We first identify and describe the different types of thoracic physiological information which EIT may be able to provide, and some capabilities and issues associated with each application. We then rate, in our opinion, the maturity of the technology, using the scale 0–10 of table 1. As an example of an established technology we may consider electrocardiography, which is also based on surface electrode measurements. In ECG monitoring, electrode types, placement, recording parameters and the test protocol are standardized. In clinical use, ECG measurements are analysed and presented in a predefined way, with a focus on extracting values relevant to clinical diagnosis and decision making. The potential failure modes of the equipment are well understood. At the same time, new research is exploring novel parameters and uses of these ECG data. We compare ECG to EIT monitoring in table 2.

Table 1. Technological maturity levels for a monitoring technology such as EIT.

Level  description
$\left.{\begin{array}{ll@{}} \ \, 0 &\hspace{10pt}{\rm Concept\ for\ technology}\\ \ \,1&\hspace{10pt}{\rm Promising\ simulation\ results}\\ \ \,2&\hspace{10pt}{\rm Phantom\ experimental\ results}\\ \ \,3&\hspace{10pt}{\rm Pilot}\ {\it in\ vivo}\ {\rm studies }\\ \ \,4&\hspace{10pt} {\rm Animal\ experiments\ validated\ to\ standards}\\ \ \,5&\hspace{10pt}{\rm Consistent\ studies\ showing\ accurate/robust\ results\quad\qquad\qquad\qquad\!\!\!\!\! }\\ \end{array}}\right\rbrace$	${\begin{array}{l@{}} {\it Goal:}\\ {\rm accurate}\\ {\rm technology}\\ {\rm (retrospective)} \end{array}}$
$\left.{\begin{array}{ll@{}} \ \,6&\hspace{7pt}{\rm Experimental\ studies\ managing\ animals\ with\ technology}\\ \ \,7&\hspace{7pt}{\rm Prospective\ studies\ managing\ patients\ with\ technology}\\ \ \,8&\hspace{7pt}{\rm Consensus\ on\ therapeutic\ use\ of\ technology}\\ \ \, 9&\hspace{7pt}{\rm Numerous/multi}\hbox{-}{\rm centre\ studies\ linking\ use\ of\ technology\ to\ outcomes\quad } \\ 10&\hspace{7pt}{\rm Routine\ clinical\ use\ and\ consensus\ on\ interpretation}\\ \end{array}}\right\rbrace$	${\begin{array}{l@{}} {\it Goal:}\\ {\rm useful}\\ {\rm technology}\\ {\rm (prospective)}\end{array}}$

Based on these levels of technological maturity, we classify and describe the fields of use which have been proposed for thoracic EIT. Table 3 is classified according to the classical categories of respiratory function (Barret et al 2009, Hall 2011, Schmidt et al 2011), and describes the respiratory function, measurement goal, capabilities and issues of each potential application of EIT. Respiratory function is separated into categories (distribution of tidal volume, intrathoracic gas volume and respiratory system mechanics) which are controlled directly by the respective ventilator parameters (tidal volume, positive end-expiratory pressure (PEEP) and lung compliance). Using this analysis, we summarize the fields of use and application maturity into two broad categories.

• Monitoring ventilation (including respiratory system mechanics and intrathoracic lung volume changes) is at a level which supports some confidence in its accuracy and repeatability. It is ready for the next step: prospective use as a tool to guide therapy (and we note that some work in this direction is underway).
• Monitoring perfusion and gas exchange is at an earlier level, but shows important promise. These applications are seeing continued experimental validation to understand their accuracy and limitations.

Many EIT systems have been developed and used for lung monitoring; most of them have been developed by academic groups with a focus on research, but recently, commercial systems have been introduced. We clarify the authors' relationships with producers of EIT equipment used in clinical measurements in table 4. The applications, capabilities and issues we discuss  are based on current EIT hardware and imaging technology, which is largely derived from the original Sheffield EIT system (Brown and Seagar 1987). In all current EIT systems designed for thoracic use, there are, as in the Sheffield system, a small number (8–32) of electrodes placed in a transverse plane around the thorax, and current is driven across electrode pairs while differential voltages are measured. Similarly, image reconstruction approaches are based on linear, one-step reconstruction algorithms, structurally similar to the Sheffield backprojection algorithm. At the same time, we also acknowledge the important progress of the last 30 years. Current EIT systems have greatly reduced noise and interference, are capable of making measurements at much higher rates (sufficient for analysis of both ventilation and perfusion), and recent image reconstruction algorithms provide improved resolution and robustness to electrode errors with reduced image artefacts (Adler et al 2009, Bodenstein et al 2009, Borges et al 2012, Costa et al 2009b).

There are several advanced EIT hardware and image reconstruction and analysis concepts being proposed which have yet to be systematically tested in reproducible technical verification environments, and validated in experimental or clinical lung EIT scenarios. This is perhaps due to an unfortunately common disconnect between the research teams in experimental and clinical lung EIT on the one hand, and their mathematical and engineering colleagues on the other.

Examples of such promising technologies are as follows. For EIT Hardware, 3D EIT systems, using multiple bands of electrodes and designed to create a 3D image of conductivity changes in a larger section of the lungs (Metherall et al 1996), and multi-frequency EIT systems, making EIT measurements at multiple stimulation frequencies to detect and image the characteristic differences between tissue types (Brown et al 1994). Lung EIT has seen little exploration of these ideas, and the majority of EIT studies have been performed using single electrode plane single-frequency systems. However, we note that such examinations provide large amounts of information to be interpreted by a clinical user.

For image reconstruction algorithms, technology examples are as follows: algorithms formulated to preserve edges (Borsic et al 2010, Rahmati et al 2011) and use robust data norms (Borsic and Adler 2009), which promise higher quality images with reduced sensitivity to data errors. Reconstruction of absolute EIT images would provide very useful data on the state of lung tissue and its absolute volume (with the promise of a Hounsfield-like unit). Unfortunately, absolute lung EIT has seen relatively little success, largely due to inexactly known electrode positions and contact impedance, and movement during breathing and posture change, even though such absolute images are routinely attempted in breast and head EIT. One promising approach is to fit a smaller number of absolute EIT parameters (such as heart and lung regions of interest (ROIs)) onto which the dynamic EIT image may be superimposed (Nebuya et al 2006).

One area of technological development which we feel shows significant promise is novel image analysis algorithms with a particular emphasis on the analysis of dynamic behaviour. The relatively high acquisition rate of modern EIT systems, and the availability of powerful temporal filtering techniques, means that small differences in the dynamic behaviour of tissues can be detected. Such algorithms allow measurement of, for example, of pressure–volume curves (Grychtol et al 2009, Kunst et al 2000, van Genderingen et al 2003), regional time constants (Pulletz et al 2009), regional opening and closing pressures (Kunst et al 2000, Pulletz et al 2011), regional respiratory system compliance (Costa et al 2009a, Dargaville et al 2010, Lowhagen et al 2010, Wolf et al 2012), ventilation delay (Wrigge et al 2008, Victorino et al 2004), regional intratidal gas distribution (Lowhagen et al 2010) and potentially recruitable lung volume (Lowhagen et al 2011). Machine learning techniques for automatic EIT image segmentation show promise (Grychtol et al 2010, Gómez-Laberge et al 2011). One important advance in such image analysis algorithms is that the EIT output to specific ventilator protocols is analysed (for example the regional opening pressure in response to slow flow inflation or PEEP trials). EIT signals corresponding to specific physiological processes may be distinguished from background events by QRS gating to measure the pulsatility in the lung regions, or to calculate the stroke volume (Vonk-Noordegraaf et al 2000, Smit et al 2003). Recently, nonlinear frequency filtering approaches were developed to determine pulse arrival times (Solà et al 2011, who fit a parameterized function to temporal signals), regional lung blood flow (Borges et al 2011, who identified the maximum slope after bolus injection) and heart–lung interactions (Maisch et al 2011). Various EIT measures can be derived from such measurements performed during specific manoeuvres, and also during continuing artificial ventilation. Not all of these measures enable the assessment of all relevant phenomena at once (e.g. alveolar recruitment, cyclic opening and closing, overdistension), and thus a combination of some indices might be useful.

While such advances promise increased clinical capabilities, and we want to encourage their development, such capabilities are not part of 'where we are today'. For current systems, we consider the following capabilities to not be sufficiently robust (primarily due to hardware limitations, such as measurement errors and drift, and unknown shape and electrode positions): (1) absolute impedance (and measures of lung volumes, in particular functional residual capacity (FRC)) which is too sensitive to the exact body shape, electrode positions, contact impedances and 3D conductivity outside of the electrode plane, and (2) prolonged comparison to a previous reference to measure slow changes in volume status over the range of hours to days, primarily due to the drift in measurement electronics, slow changes in electrode–skin contact impedance and body shape changes due to postural shifts.

However, despite the limitations of today's technology (at the hardware and image reconstruction level), there is clear experimental evidence that current EIT systems are of sufficient quality for clinically useful lung measurements, and especially for monitoring of ventilation. In the remainder of this paper, we therefore focus on EIT for ventilation monitoring. Here, EIT is ready to go from retrospective analysis of its accuracy to prospective guidance for patient therapy. We are ready to move from asking, 'Can we measure?' to 'Can we guide therapy?'.

Research must be focused on relevant outcome variables. First, it is necessary to find adequate interim surrogate outcomes that use EIT.  For example, the focus can be on outcomes like (a) shorter time to achieve clinical treatment goals, such as improved oxygenation index; (b) less adverse events, such as pneumothorax (Costa et al 2008, Miedema et al 2011); (c) less VILI, resulting in reduced markers of inflammation. Many possible EIT measures have been proposed, such as EIT measures of spatial homogeneity of ventilation distribution using centre of ventilation, (Karsten et al 2011, Frerichs et al 1998), ventilation profiles (Frerichs et al 2006), regional opening and closing pressures (Kunst et al 2000, Pulletz et al 2011), lung filling dynamics (Hahn et al 1996, Kunst et al 2000, Riedel et al 2009, Hinz et al 2007, Miedema et al 2011, Wrigge et al 2008, Frerichs et al 2001, Pham et al 2011), respiratory system compliance (Costa et al 2009a, Dargaville et al 2010) and intratidal regional gas distribution (Lowhagen et al 2010). At the present time, we cannot yet investigate mortality and morbidity variables, since these need carefully structured multi-centre trials, and we do not know how to best use EIT in the clinical setting. However, it is essential that EIT be measured against its ability to improve mortality and morbidity rates in order for EIT to become part of routine clinical practice.

We are thus arguing that clinical research studies should focus on whether EIT can be used to measure variables leading to relevant diagnoses (e.g. hyperinflation). These diagnoses would be used to guide patient therapy. Since such EIT-guided therapy is new and unproven, there will be, reasonably, an initial reluctance to give research ethics permission. This argues for a need for animal experiments in which EIT-guided therapy is used to treat experimental models of lung injury.

To clarify the clinical research, we give a few examples of research projects which explore interesting, achievable and relevant questions.

• EIT-guided recruitment (for treatment of atelectasis). Hypothesis: EIT-guided recruitment manoeuvres followed by EIT-guided determination of optimal PEEP results in better gas exchange and reduced inflammation. In this case recruitment and PEEP would be selected to achieve the best reduction in EIT-measured fractional reversible atelectasis without a large increase in overdistension.
• Fast feedback (for guiding care). Hypothesis: EIT can give feedback in seconds on changes in regional lung function, and this rapid feedback can make faster, safer and better optimized control of ventilation. For example, studies could investigate whether the selected PEEP is different in patients managed with and without EIT.
• Warnings (quality and safety). Hypothesis: EIT can provide early warnings about sub-optimal ventilation or dangerous conditions. In this case, studies could investigate: (1) During changes in treatment, is the time course of physiological variables different with or without EIT monitoring? (2) With or without an EIT system to warn for pneumothoraces, are the number of events and pattern of care different? (3) Can EIT reliably detect pneumothorax in patients (and validate against x-ray)?
• Clinically relevant diagnostic EIT measures (for guiding ventilator settings). Hypothesis: EIT can identify relevant deleterious and favourable phenomena (like cyclic opening and closing of lung units and overdistension on one hand and alveolar recruitment on the other). Studies could compare different EIT indices against each other, for measure of both overdistension and cyclic opening and closing. This would allow determination if patients who require a special manoeuvre (for example a slow inflation) yield comparable measures to those available during ongoing therapeutic ventilation. Also, it should be determined if they can be used in both volume- and pressure-controlled mechanical ventilation.

Much of this work could be facilitated by the creation of an on-line database of clinical EIT data, which would enhance collaboration and further research, especially by helping EIT groups without access to such data (to develop, for example, more accurate and robust algorithms). Data that have been made available through eidors.org have already gone a long way to facilitating this collaboration.

In this section, we focus on achievable, practical engineering improvements that relate to the current generation of EIT hardware and algorithms. These recommendations are meant to be for both the academic engineering researchers and for companies developing EIT products.

The most pressing EIT hardware need is for the equipment to be available in sufficient quantities at a reasonable cost. Addressing the practical problems associated with electrodes is the most important requirement for image quality. We need to understand what electrode types, shapes, supports and contact fluid or gel work best with which patients (with also a consideration to placement on preterm infants). Next, there is a requirement for improved patient interfaces which allow EIT electrodes to be placed quickly, reliably and easily. The electrode support structures/belts recently introduced help address this problem, although they need to be placed with a fairly large tension to achieve adequate stability and electrode contact, which may reduce chest wall compliance.

EIT systems can be optimized in several ways. Most lung EIT systems use stimulation patterns via adjacent or nearby electrodes, which are sub-optimal (Adler et al 2011); modifications of these patterns should yield improved signal robustness. Multi-frequency EIT promises useful clinical information on tissue types which could make EIT measures more robust. For example, it may be possible to identify oedematous lung tissue or to distinguish between collapsed (non-ventilated) lung and other tissues. Finally, current configurations place electrodes in a single plane which cannot adequately image 3D distributions of lung ventilation. It is, of course, possible to create 3D images with multiple electrode belts, but the benefit has not been clarified: that is,what specific imaging advantages are available from two or three electrode planes (Bikker et al 2011, Frerichs et al 1999, Reifferscheid et al 2011), and is this information worth the inconvenience and cost of placing extra belts?

There is now a general consensus on the requirements of linear reconstruction algorithms for lung EIT (Adler et al 2009). In addition to these requirements, it is important that the shape of the model domain be roughly right (not simply a circle) and that the very different conductivity of the lungs (versus most other tissues) be accounted for (Grychtol et al 2011). However, given the consensus on the linear algorithms, there are several other aspects of algorithm behaviour which will need to be validated and tested against models, phantoms or experiments. An understanding of reliable ways to automatically and reproducibly select ROI in lung images (Pulletz et al 2006) will need to be developed. As approaches to analyse regional information from EIT images are developed, it will be important to understand and compensate for the partial volume effects (blurring between regions) caused by the limited resolution of the reconstruction algorithm. Similarly, the practical extent to which out-of-electrode-plane conductivity changes affect EIT images is not yet clear. While absolute EIT is not currently sufficiently robust, there is a strong case for using the estimated absolute properties to improve the lung volume measurements. It is common to use frequency domain filtering to select or remove ventilation and cardiac frequencies (Carlisle et al 2010, Dunlop et al 2006, Fagerberg et al 2009, Frerichs et al 2009), and filtering may be enhanced using linear transforms (Deibele et al 2008) or a second level of parametric models (Solà et al 2011). Since such filtering loses information, and physiological frequencies (especially cardiac) are not constant, it is important to develop guidance to determine when a filter can lose valuable impedance change information.

Physiological parameters calculated from EIT have largely been based on the change in impedance between end-expiration and end-inspiration. This parameter correlates closely to regional volume changes, and can thus be used as a measure of compliance, provided the driving pressure is known. There is rich information available in the dynamics of the EIT signals, and we encourage the work to develop parameters to exploit this information, which should yield opening/closing pressures, time constants and measures of overdistension, collapse and tidal recruitment. Finally, to interface with other medical imaging research and hospital data systems, it must be possible to write EIT images to a standard format. This implies a need to define a DICOM storage class for EIT images.

Clinical EIT systems require usable software to interact with. The user interface must be designed with such clinical interactions and use in mind. Displayed parameters must be easy to interpret and preferably be in physiological units (Lundin and Stenqvist 2012). It is also essential that EIT systems control the quality of data and help the users to correctly interface to patients. Specifically, any errors in electrode contact or incorrect placement should be detected and, if possible, automatically corrected, to avoid the resulting serious image artefacts. The importance of automatic quality control and decision support is underlined by a recent study, showing that the ability of clinicians to detect patient-ventilator asynchronies is low (Colombo et al 2011).

EIT images and data need to be represented consistently in a clinically useful way. The choice of colours for representing images needs to be consistent. For example, if a set offset to represent FRC is added to the images, it could be misleading unless it were clear that it were just for illustrative purposes. As a sample proposal, a colour map used by Dräger Medical has been made available (on eidors.org). As more EIT information is simultaneously represented (eg. opening and closing pressures, overdistension and collapse) the consistency of image colour choices will become more important. Another aspect of standardization is the requirement to choose a reference dataset for time difference imaging. In some cases the average signal is chosen, while in others, end-expiratory events are identified and used as reference. The requirement for a reference dataset can be confusing to EIT users, and it would be useful to have a consistent way to indicate how it is done.